Insights  /  Case study

Case Study · Payment Integrity

Roughly $300K a year was leaking out before payroll. We built the gate that stops it.

A licensed home-care agency was validating caregiver hours by hand, across a wall of spreadsheets. Our team replaced it with a time-aware control system that checks every hour against what was actually authorized, before a single dollar is paid.

Prevented / month

$24–30K

Leakage & clawbacks caught before payout

Payroll QA cycle

8h → <2h

~75% less clerk time, every run

Recovered / year

~$300K

Estimated, at current volume

Enforcement

6 / 6

Validation rules run on every visit, no skips

The situation

Payroll was a manual reconciliation against spreadsheets no one fully trusted.

Every pay period, a clerk sat down with progress notes on one side and a sprawl of authorization spreadsheets on the other, and reconciled them by hand. Were these hours inside the certification period? Was this the assigned caregiver on that date? Did the week exceed the authorized limit? Had this visit already been paid? Was the client on hold?

Each of those questions is a place money leaks: an over-authorized week, a duplicate paid twice, a visit outside the cert period that triggers a clawback later. Answered by hand, across eight hours, at the end of a long month, mistakes were inevitable and expensive. The spreadsheets couldn't answer "what was true on the service date", only what was true today.

What we built

A control system, not a report. It decides what is eligible to be paid, and refuses the rest.

The system sits upstream of payroll as a gatekeeper. Every hour of care is checked against the authorization that governed it on the date it was delivered. Anything that fails is held and surfaced for review, with the reason attached.

i.

Time-aware authorization

Every check uses the rules that were in force on the service date, not today's. Past pay periods always validate against what was actually true then, so late submissions and back-dated corrections stay correct.

ii.

Six rules, every visit, no shortcuts

Cert period, correct caregiver, authorized day, weekly-hour ceiling, active holds, and duplicate payments. All six run on every visit record. Blocks are hard; nothing slips through on a busy day.

iii.

Document intake that reads itself

Scanned progress notes are read by AI into structured visit records. Low-confidence fields are flagged for a clerk to confirm against the image. Submission is blocked until they do. Speed without blind trust.

iv.

Compliance & recertification, tracked

Certification windows and holds are first-class. The system flags what's expiring before it lapses, so authorizations don't quietly run out and turn billable work into unpaid work.

The checks

Six checks. Six ways money used to walk out.

Every visit is tested against all six, every run. No shortcuts, no skips. Five block a payout outright; one flags a visit for a human to look at. Each maps to a specific way dollars used to escape.

Duplicate

The same visit, paid twice

No hour is paid if a record already exists for that patient, date, and care type. Catches the double-entry that manual review misses at hour seven.

Blocks payout
Weekly ceiling

An over-authorized week you'll claw back later

Hours are summed across the true Sunday–Saturday week and held the moment they exceed the authorized limit, before the overage is ever paid.

Blocks payout
Cert period

Care delivered outside the authorized window

A service date beyond the certification period is stopped, the classic source of denied claims and after-the-fact recoupment.

Blocks payout
Hold status

Paying for care during a hospitalization or hold

If the client was on an active hold on the service date, the visit doesn't pass. Those hours aren't reimbursable and shouldn't reach payroll.

Blocks payout
Aide assignment

The wrong caregiver on the wrong day

The hours must belong to the caregiver actually assigned on that service date. Unassigned work is unbillable work, caught before it's paid.

Blocks payout
Authorized day

A visit on a day that wasn't authorized

Visits landing outside the authorized days of the week are surfaced for a clerk to confirm. A flag, not a hard stop, because these are sometimes legitimate.

Flags for review

Before & after

Same payroll run, two different operations.

Before · by hand

  • ×8 hours per pay period, one clerk cross-referencing spreadsheets
  • ×Errors surface weeks later as clawbacks and denied claims
  • ×No reliable answer to "what was authorized on that date"
  • ×Duplicates and over-authorized weeks slip through under time pressure
  • ×Recertifications tracked in someone's head

After · the control system

  • Under 2 hours, reviewing only what the system flagged
  • Invalid hours stopped before payout, not recovered after
  • Point-in-time truth for every service date, always
  • Duplicates and ceiling breaches blocked automatically
  • Expiring authorizations flagged before they lapse
The win isn't that payroll got faster. It's that money which used to walk out the door, quietly, in ones and twos, every single run, now doesn't.The outcome, in one line

Why this generalizes

Home care was the proving ground. The pattern is anywhere money moves against rules.

Strip away the domain and the shape is universal: high-volume work, validated by hand against authorizations and policy, right before money changes hands. Wherever that describes an operation, the same control layer recovers leakage and collapses the review cycle.

The work
Any recurring, high-volume claim of value: hours, visits, invoices, reimbursements, insurance claims, contractor billing.
The rules
Authorizations, contracts, eligibility, policy ceilings, the constraints that determine what is actually owed.
The gate
Automated validation before payout, evaluated against what was true on the date the work was done.
The return
Recovered leakage, fewer clawbacks, an audit trail, and hours of manual review returned to the team every cycle.

The same gate, for insurers

Claims adjudication, medical-billing QA, and payment integrity are the exact same problem: validate each claim of value against the policy, authorization, and history that governed it, before money leaves. OKEMA brings this event-based, point-in-time control approach to insurers, banks, and telecoms.

See Claims Intelligence →    See Fraud Intelligence →

Where is your operation leaking?

A focused working session: we map where money escapes between work-done and paid, and size what a control layer would recover on your book.

Book an AI Opportunity Call →